Sockets

Overview

A socket is an endpoint for sending or receiving data across a computer network. It’s the fundamental programming interface for network communication, providing a file descriptor abstraction for network I/O.

What is a Socket?

graph LR
    subgraph "Process A"
        SA[Socket A<br>IP: 10.0.0.1, Port: 52341]
    end
    subgraph "Process B"
        SB[Socket B<br>IP: 10.0.0.2, Port: 80]
    end
    SA <-->|TCP/UDP Connection| SB

A socket is defined by a 5-tuple:

(source IP, source port, destination IP, destination port, protocol)

Socket Types

Socket API (Berkeley Sockets)

TCP Socket Lifecycle

sequenceDiagram
    participant S as Server
    participant C as Client
    S->>S: socket()
    S->>S: bind(IP, port)
    S->>S: listen()
    C->>C: socket()
    C->>S: connect() → SYN
    S->>C: SYN-ACK
    C->>S: ACK
    S->>S: accept() → new socket
    C->>S: send() → data
    S->>C: recv() → data
    C->>C: close() → FIN
    S->>S: close() → FIN

UDP Socket Lifecycle

sequenceDiagram
    participant S as Server
    participant C as Client
    S->>S: socket(SOCK_DGRAM)
    S->>S: bind(IP, port)
    C->>C: socket(SOCK_DGRAM)
    C->>S: sendto(data)
    S->>C: recvfrom(data)
    C->>S: sendto(data)
    S->>C: recvfrom(data)

Socket States

Socket Options

Socket Buffer

graph LR
    subgraph "Send Buffer"
        APP[Application writes] --> SB[Socket Send Buffer]
        SB --> NET[Kernel sends to network]
    end
    subgraph "Receive Buffer"
        NET2[Network data arrives] --> RB[Socket Receive Buffer]
        RB --> APP2[Application reads]
    end

When the send buffer is full, send() blocks (or returns EAGAIN for non-blocking). When the receive buffer is full, TCP advertises a smaller window (flow control).

Interview Questions

1. Q: What is a socket? A: A socket is an endpoint for network communication, identified by (IP, port, protocol). It provides a file descriptor abstraction for sending/receiving data over the network. The Berkeley Sockets API (socket, bind, listen, accept, connect, send, recv) is the standard interface.

2. Q: What’s the difference between socket, bind, listen, and accept? A: socket() creates a socket. bind() assigns an address (IP+port) to it. listen() marks it as passive (server socket). accept() waits for and returns a new socket for each incoming connection.

3. Q: Why does a server need both listen() and accept()? A: listen() creates a listening socket that queues incoming connections. accept() dequeues a connection and returns a new socket for communicating with that specific client. The listening socket remains open for new connections.

4. Q: What is SO_REUSEADDR? A: Allows binding to an address that’s in TIME_WAIT state. Without it, restarting a server may fail with “Address already in use” because the old socket is still in TIME_WAIT. Essential for server applications.

5. Q: What is TCP_NODELAY? A: Disables Nagle’s algorithm, which buffers small writes into larger TCP segments. With TCP_NODELAY, data is sent immediately. Important for latency-sensitive applications (gaming, SSH, real-time systems).

Common Mistakes

• Forgetting to call bind() before listen() (OS assigns random port)
• Not setting SO_REUSEADDR (can’t restart server quickly)
• Not handling partial reads/writes (send/recv may return less than requested)
• Using blocking sockets in single-threaded servers (blocks on one client)
• Closing the listening socket instead of the accepted socket

Summary

Sockets are the foundation of network programming. Understanding the socket lifecycle (socket → bind → listen → accept for servers, socket → connect for clients), socket options (SO_REUSEADDR, TCP_NODELAY), and buffer management is essential.

Cross-References

• TCP Sockets
• UDP Sockets
• Unix Domain Sockets
• Non-blocking I/O
• I/O Multiplexing

Cross References

• TCP Sockets
• UDP Sockets
• Unix Sockets
• I/O Multiplexing
• OS IPC Sockets